Kuaishou's Kling AI released Kling Image 4.0 on November 17. It's a still-image model, separate from the Kling video models, and it's live in Kling's web studio and through its API (Kling AI, Pandaily). Kling is pitching three things: native 8K output, up to 2,000 characters of text in one image, and sets of up to 10 related images from a single prompt.

The launch post on Kling's X account had about 1,250 likes and just over a million views by November 19 (X). Venice, an AI platform, added the model the same day (X).

This post covers what Kling announced, which claims have been checked so far, what the model costs, and three prompts that borrow its two most useful ideas: a whole image set from one prompt, and lots of exact text. We ran them on GPT Image 2 and Nano Banana on EasyImagen.

What Kling announced

From Kling's product post and the launch press release:

  • One model for understanding and generating. Kling says Image 4.0 handles both in a single autoregressive framework, and says this helps it turn a detailed creative brief into finished images.
  • Native 8K. Kling calls it an industry first. It hasn't published exact pixel sizes. 8K usually means 7,680 pixels on the long side, about 33 megapixels at 16:9.
  • Up to 2,000 characters of text per image, with better small text. Prompts can be up to 15,000 tokens. Pandaily points out that these are two different limits: a long prompt doesn't guarantee that all its text appears correctly in the image.
  • Typography controls. There are more than 70 preset fonts, you can upload a font reference image, and you can edit text and font style in an image after it's generated. Text works in English, Japanese, Korean, Spanish, Portuguese, and simplified and traditional Chinese.
  • Sets of up to 10 images from one prompt. Kling says the model plans the whole set, decides what each image shows, and keeps the style and subject consistent. It suggests storyboards, product displays and presentation slides. A separate long-image mode makes e-commerce detail pages and web layouts.
  • Color grading. There are more than 110 preset color-grading styles, and you can copy the color and lighting of a reference image.
  • Varied group portraits. Kling says crowds get a wider range of faces, body types, ages and poses, instead of the usual identical people.

What's checked, and what it costs

Rankings. The independent Artificial Analysis leaderboard ranks models by blind votes between pairs of images. As of November 18, OrcaRouter reported Kling Image 4.0 in 7th place for text-to-image (Elo 1,180) and 6th for image editing (1,183). The previous Kling Image 3.0 Omni ranked 87th, about 255 points lower (OrcaRouter). That's a big jump, but Kling isn't at the top: GPT Image 2.6 Sunburst leads text-to-image at 1,261 (Fello AI).

Unverified claims. No public benchmark measures resolution, so "native 8K" is Kling's own claim for now. The 2,000-character text limit hasn't been tested independently either.

Price. On Kling's developer price list, a 2K image costs 16 units, 4K costs 24 and 8K costs 40. At the entry rate of about $0.0038 per unit, that's roughly 6, 9 and 15 cents per image (OrcaRouter). Reference images are now billed too, at 0.8 units each, where earlier Kling image models didn't charge for them. If you build sets from several references, factor that in.

Does 8K matter? Mostly for print. At 300 dpi, a 7,680-pixel side covers about 25.6 inches (65 cm). On a web page, an image is rarely shown wider than 1,600 pixels. Our MiniMax H3-Image post has more on when high resolution actually matters.

Prompting a set: a storyboard in one image

Kling's set feature plans several images at once. On models without that feature, you can get a similar result by asking for one sheet with a numbered grid of frames. The trick is the same one Kling describes: describe the character once and in full, then give each frame one line about what it shows.

This is the cover image, made with GPT Image 2 at 16:9:

A six-panel storyboard sheet for a short animated story, laid out as a 3 by 2 grid of equal 16:9 frames with thin white gutters on a charcoal background, each frame numbered 1 to 6 in a small white number in its top-left corner. The same main character appears in every frame and must look identical each time: a small red fox with a cream chest, a green knitted scarf and a tiny brass lantern held in its mouth or paws. Frame 1: wide shot of a snowy hillside village at night, the fox stepping out of a burrow under a pine tree, the lantern glowing. Frame 2: medium shot of the fox trotting along a snowy lane between half-timbered cottages with lit windows. Frame 3: close-up of the fox's face, ears back, looking up at a dark cottage where one window has no light. Frame 4: the fox climbing a woodpile toward that dark window, snow falling. Frame 5: over-the-shoulder shot of the fox setting the lantern on the windowsill as warm light spills into the room. Frame 6: wide shot of the whole village from the hill, every window now glowing, the fox sitting in the snow watching. Consistent soft painterly animation style, cold blue night palette with warm amber lantern light, consistent lighting and character design across all six frames. No other text, no captions, no logos.

The fox kept the same scarf, lantern and markings in all six frames, and the frames follow the story in order. The light also follows the story: the dark window in frame 3 is lit by frame 5. One thing to watch is that the frames come out small. For a real storyboard, generate the sheet at a large size, or generate each frame separately and paste the character description into every prompt.

Lots of exact text: a cafe menu

Kling claims up to 2,000 characters of text per image. We tested a smaller, more common case: a full cafe menu with 20 separate pieces of text, about 420 characters in all, each with a price. This ran on GPT Image 2 at 2:3:

A printed cafe menu poster on warm off-white paper, portrait layout, clean modern typography with a dark green and terracotta color scheme and a small line drawing of a steaming cup at the top. Title at the top exactly: "Hearth & Crumb". Under it, small caps exactly: "WINTER MENU · SERVED DAILY 7 AM – 4 PM". Then three sections, each with a heading and items with prices right-aligned on dotted leader lines.
Heading "COFFEE": "Espresso 3.00", "Flat White 4.50", "Oat Milk Latte 5.25", "Maple Cinnamon Latte 5.75", "Cold Brew 4.75".
Heading "TEA & COCOA": "Earl Grey 3.50", "Spiced Chai 4.75", "Peppermint Hot Cocoa 5.00", "Matcha Latte 5.50".
Heading "FROM THE OVEN": "Butter Croissant 3.75", "Cardamom Bun 4.25", "Gingerbread Loaf Slice 4.00", "Cheddar Chive Scone 4.50", "Cranberry Orange Muffin 3.95".
At the bottom, a small italic line exactly: "Ask about our gluten-free bakes. Oat, almond and soy milk +0.75". Generous margins, every word spelled exactly as written, crisp and legible small text, flat printed graphic, no other text, no logos.

A printed cafe menu titled Hearth & Crumb, Winter Menu, served daily 7 AM to 4 PM, with Coffee, Tea & Cocoa and From the Oven sections listing fourteen items with prices on dotted leader lines, and a footer about gluten-free bakes and milk alternatives

All 20 pieces of text came out spelled correctly, in the right order, and with the right prices. Three habits made that work, and they apply to any model:

  1. Put every piece of text in quotes, exactly as it should look, prices included.
  2. Group the text under headings in the prompt the same way it should be grouped on the page.
  3. Describe the layout, such as "prices right-aligned on dotted leader lines", so the model doesn't have to invent where text goes.

The same approach drives our printable pages, such as the cookie exchange invitation and this week's Thanksgiving place cards prompt.

A product set: four shots of one mug

Kling suggests image sets for product listings. Here is a four-photo listing set in one image, made with Nano Banana at 1:1. As with the storyboard, describe the product once and in detail, then give each shot one line:

A 2 by 2 grid of four product photos for an online shop listing, all showing the exact same handmade ceramic mug: a rounded stoneware mug with a speckled oatmeal-colored glaze, a deep blue glaze dripping down from the rim about a third of the way, an unglazed sandy foot ring and a C-shaped handle. Top left: the mug alone on a seamless pure white background, front view, soft even studio light. Top right: lifestyle shot of the mug filled with black coffee on a light oak table next to an open book and a linen napkin, morning window light. Bottom left: extreme close-up of the blue glaze drip meeting the speckled glaze, showing texture and tiny pinholes. Bottom right: the mug held in two hands wearing a cream knit sweater sleeve, cropped at the wrists, to show its size. Same mug shape, glaze colors and drip pattern in all four photos, consistent color grading, thin white gutters between the photos. No text, no logos, no faces.

Four product photos of the same speckled stoneware mug with a blue glaze drip from the rim: alone on a white background, filled with coffee on an oak table next to an open book, a close-up of the glaze texture, and held in a hand with a cream knit sleeve

The mug's shape, glazes and unglazed foot match in all four shots, and the lighting changes the way it would between a studio shot and a lifestyle shot. The drip pattern isn't identical, though: the number and length of the blue drips change slightly from shot to shot. That doesn't matter for a mood board, but it would for a real listing, where buyers expect every photo to show the same item. For that, edit photos of the real product instead of generating new ones.

Try it now

Kling Image 4.0 isn't on EasyImagen. GPT Image 2, Nano Banana and Seedream 5 Pro are, so you can run these prompts and compare the models in the generator. For a set, describe the subject once, then write one line per frame or shot. For text, put every word in quotes. For more seasonal ideas, try the Christmas cookie generator.